Stockfish 2.2 JA Time management.

[mcostalba]

Thanks to everybody for reporting these issues.

To turn these reports in something useful to identify and fix the problem I'd ask to the people reporting time management problems to post following information:

1) Operating system used: Windows / Linux, 32 or 64 bit

2) Number of threads used

3) Time control used

Thanks a lot for your support.

Marco

[IWB]

IPON Conditions


1) Operating system used: Windows / Linux, 32 or 64 bit

XP64

2) Number of threads used

ONE

3) Time control used

5 + 3

Right now, after 2000+ games it is somewhere between 2-3% - which is very high compared to others. Exact numbers after I finished the run.

Bye
Ingo

PS: For what its worth, but at the german CSS fora I read from S. Pohl that another windows compile of the original MAC Stockfish 2.2 runs fine with ultra short time controls ...

[pohl4711]

On my old Intel Quad (no SSE support, Vista 64bit) Stockfish 2.2 JA 64bit looses 1/3 of all games on time under LittleBlitzerGUI with 60 seconds + 500ms Fischer Bonus. Version 2.1.1 doesnt...
Testing for my NEBB-Rankinglists is impossible...

Stefan

Hi again,

I tried the Stockfish 2.2 64bit compile of PeterPan. This one works fine and I got no more time losses under the LittleBlitzerGUI.
Testrun for my NEBB-rankinglists started now.

Her the download-link of the PeterPan-Compile (64bit only, not working with Windows xp, only Vista and 7):
http://www.mediafire.com/?y1x2r49ch7g251u

[pohl4711]

Thanks to everybody for reporting these issues.

To turn these reports in something useful to identify and fix the problem I'd ask to the people reporting time management problems to post following information:

1) Operating system used: Windows / Linux, 32 or 64 bit

2) Number of threads used

3) Time control used

Thanks a lot for your support.

Marco

Hi Marco:
1) Vista 64bit Home Premium
2) 1 Thread, no ponder, 4 games same time on Quadcore with LittleBlitzerGUI
3) 60 seconds per game + 500ms Fischer-Bonus

With Stockfish 2.1.1 not a single time loss in 900 games with the same conditions, Stockfish 2.2 JA 64bit 35-40% time losses (!!!) of all games...

Best - Stefan

P.S: The 2.2 PeterPan-Compile works fine (see my posting above)...

[MM]

Thanks to everybody for reporting these issues.

To turn these reports in something useful to identify and fix the problem I'd ask to the people reporting time management problems to post following information:

1) Operating system used: Windows / Linux, 32 or 64 bit

2) Number of threads used

3) Time control used

Thanks a lot for your support.

Marco

Hello Mr Costalba,


windows 7 64 bit
1 single core used for both, ponder ON
2 different TC: 1'+0s per game and 10 seconds per game.

p.s. i think it is a problem especially for fast games without increment, my experience is a huge number of blunders in time trouble and a lot of time losses.

(Fritz 13 GUI)

Best Regards

[mcostalba]

Her the download-link of the PeterPan-Compile (64bit only, not working with Windows xp, only Vista and 7)

Thanks everybody. This is an important hint.

Actually the JA compiles use a so called OLD_LOCKS configuration to be backward compatible with XP. Instead for Vista and 7 there is no need of this and SF can use the new (and more appropriate) SRW locks (that are also the default ones when compiling for Windows).

Jim will try to do a version with SRW locks (only for Vista and Windows 7) in the next days, so we will actually verify if the problem is this.

Thanks
Marco

[MM]

Her the download-link of the PeterPan-Compile (64bit only, not working with Windows xp, only Vista and 7)

Thanks everybody. This is an important hint.

Actually the JA compiles use a so called OLD_LOCKS configuration to be backward compatible with XP. Instead for Vista and 7 there is no need of this and SF can use the new (and more appropriate) SRW locks (that are also the default ones when compiling for Windows).

Jim will try to do a version with SRW locks (only for Vista and Windows 7) in the next days, so we will actually verify if the problem is this.

Thanks
Marco

Thank you,

in my test 1'+1s i had no problems with 2.2 (cause the increment i think). I published the results in this forum...if you are interested....

http://talkchess.com/forum/viewtopic.php?t=41725

Best Regards

[IWB]

Hi Marco,

Every other Engine runs on W7/Vista and XP (and other Windows) without a timing problem ... I have some doubts that this is the solution. Besides that, I am running on XP64 (which is a W-Server 2003) and I am having problems too ... maybe just not that "heavy" as 3 seconds increment are a lot ...

Bye
Ingo

[LucenaTheLucid]

I've had 2 time losses in probably 60 games of 40/40

Windows 7 64-bit
1 thread
40 moves/40 minutes

[mcostalba]

Hi Marco,

Every other Engine runs on W7/Vista and XP (and other Windows) without a timing problem ... I have some doubts that this is the solution. Besides that, I am running on XP64 (which is a W-Server 2003) and I am having problems too ... maybe just not that "heavy" as 3 seconds increment are a lot ...

Bye
Ingo

Hi Ingo,

yes you are right, SF should work without time losses also on XP.

Probably I have found the bug, as I suspect in the OLD_LOCKS code. But before to publish the new version I'd need to be sure no more time losses are presents. So I ask for someone that could test the new compile just to verify no more time lost under _any_ time control. If someone is interested please drop me a pm writing me the e-mail where to send the new package. It would be optimal that is tested by the same people that has experienced time management issues so to verify that under the same conditions the problem is fixed.

Thanks
Marco